kernel32/tests: Add tabular UTF-7 decoding tests.
This commit is contained in:
parent
6eab498160
commit
ddc0c3203f
|
@ -681,6 +681,137 @@ static void test_utf7_decoding(void)
|
|||
41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, -1, -1, -1, -1, -1 /* 0x70-0x7F */
|
||||
};
|
||||
|
||||
struct
|
||||
{
|
||||
/* inputs */
|
||||
char src[32];
|
||||
int srclen;
|
||||
WCHAR *dst;
|
||||
int dstlen;
|
||||
/* expected outputs */
|
||||
WCHAR expected_dst[32];
|
||||
int chars_written;
|
||||
int len;
|
||||
}
|
||||
tests[] =
|
||||
{
|
||||
/* tests string conversion with srclen=-1 */
|
||||
{
|
||||
"+T2BZfQ-", -1, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{0x4F60,0x597D,0}, 3, 3
|
||||
},
|
||||
/* tests string conversion with srclen=-2 */
|
||||
{
|
||||
"+T2BZfQ-", -2, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{0x4F60,0x597D,0}, 3, 3
|
||||
},
|
||||
/* tests string conversion with dstlen=strlen(expected_dst) */
|
||||
{
|
||||
"+T2BZfQ-", -1, output, 2,
|
||||
{0x4F60,0x597D}, 2, 0
|
||||
},
|
||||
/* tests string conversion with dstlen=strlen(expected_dst)+1 */
|
||||
{
|
||||
"+T2BZfQ-", -1, output, 3,
|
||||
{0x4F60,0x597D,0}, 3, 3
|
||||
},
|
||||
/* tests string conversion with dstlen=strlen(expected_dst)+2 */
|
||||
{
|
||||
"+T2BZfQ-", -1, output, 4,
|
||||
{0x4F60,0x597D,0}, 3, 3
|
||||
},
|
||||
/* tests dry run with dst=NULL and dstlen=0 */
|
||||
{
|
||||
"+T2BZfQ-", -1, NULL, 0,
|
||||
{}, 0, 3
|
||||
},
|
||||
/* tests dry run with dst!=NULL and dstlen=0 */
|
||||
{
|
||||
"+T2BZfQ-", -1, output, 0,
|
||||
{}, 0, 3
|
||||
},
|
||||
/* tests ill-formed UTF-7: 6 bits, not enough for a byte pair */
|
||||
{
|
||||
"+T-+T-+T-hello", -1, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{'h','e','l','l','o',0}, 6, 6
|
||||
},
|
||||
/* tests ill-formed UTF-7: 12 bits, not enough for a byte pair */
|
||||
{
|
||||
"+T2-+T2-+T2-hello", -1, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{'h','e','l','l','o',0}, 6, 6
|
||||
},
|
||||
/* tests ill-formed UTF-7: 18 bits, not a multiple of 16 and the last bit is a 1 */
|
||||
{
|
||||
"+T2B-+T2B-+T2B-hello", -1, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{0x4F60,0x4F60,0x4F60,'h','e','l','l','o',0}, 9, 9
|
||||
},
|
||||
/* tests ill-formed UTF-7: 24 bits, a multiple of 8 but not a multiple of 16 */
|
||||
{
|
||||
"+T2BZ-+T2BZ-+T2BZ-hello", -1, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{0x4F60,0x4F60,0x4F60,'h','e','l','l','o',0}, 9, 9
|
||||
},
|
||||
/* tests UTF-7 followed by characters that should be encoded but aren't */
|
||||
{
|
||||
"+T2BZ-\x82\xFE", -1, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{0x4F60,0x0082,0x00FE,0}, 4, 4
|
||||
},
|
||||
/* tests srclen > strlen(src) */
|
||||
{
|
||||
"a\0b", 4, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{'a',0,'b',0}, 4, 4
|
||||
},
|
||||
/* tests srclen < strlen(src) outside of a UTF-7 sequence */
|
||||
{
|
||||
"hello", 2, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{'h','e'}, 2, 2
|
||||
},
|
||||
/* tests srclen < strlen(src) inside of a UTF-7 sequence */
|
||||
{
|
||||
"+T2BZfQ-", 4, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{0x4F60}, 1, 1
|
||||
},
|
||||
/* tests srclen < strlen(src) right at the beginning of a UTF-7 sequence */
|
||||
{
|
||||
"hi+T2A-", 3, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{'h','i'}, 2, 2
|
||||
},
|
||||
/* tests srclen < strlen(src) right at the end of a UTF-7 sequence */
|
||||
{
|
||||
"+T2A-hi", 5, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{0x4F60}, 1, 1
|
||||
},
|
||||
/* tests srclen < strlen(src) at the beginning of an escaped + sign */
|
||||
{
|
||||
"hi+-", 3, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{'h','i'}, 2, 2
|
||||
},
|
||||
/* tests srclen < strlen(src) at the end of an escaped + sign */
|
||||
{
|
||||
"+-hi", 2, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{'+'}, 1, 1
|
||||
},
|
||||
/* tests len=0 but no error */
|
||||
{
|
||||
"+", 1, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{}, 0, 0
|
||||
},
|
||||
/* tests a single null char */
|
||||
{
|
||||
"", -1, output, sizeof(output) / sizeof(WCHAR) - 1,
|
||||
{0}, 1, 1
|
||||
},
|
||||
/* tests a buffer that runs out while not decoding a UTF-7 sequence */
|
||||
{
|
||||
"hello", -1, output, 2,
|
||||
{'h','e'}, 2, 0
|
||||
},
|
||||
/* tests a buffer that runs out in the middle of decoding a UTF-7 sequence */
|
||||
{
|
||||
"+T2BZfQ-", -1, output, 1,
|
||||
{0x4F60}, 1, 0
|
||||
}
|
||||
};
|
||||
|
||||
if (MultiByteToWideChar(CP_UTF7, 0, "foobar", -1, NULL, 0) == 0 &&
|
||||
GetLastError() == ERROR_CALL_NOT_IMPLEMENTED)
|
||||
{
|
||||
|
@ -781,6 +912,34 @@ static void test_utf7_decoding(void)
|
|||
"i=0x%02x: expected output=%s, got output=%s\n",
|
||||
i, wine_dbgstr_wn(expected, expected_len + 1), wine_dbgstr_wn(output, expected_len + 1));
|
||||
}
|
||||
|
||||
for (i = 0; i < sizeof(tests) / sizeof(tests[0]); i++)
|
||||
{
|
||||
memset(output, 0x23, sizeof(output) - sizeof(WCHAR));
|
||||
output[sizeof(output) / sizeof(WCHAR) - 1] = 0;
|
||||
SetLastError(0xdeadbeef);
|
||||
|
||||
len = MultiByteToWideChar(CP_UTF7, 0, tests[i].src, tests[i].srclen,
|
||||
tests[i].dst, tests[i].dstlen);
|
||||
|
||||
tests[i].expected_dst[tests[i].chars_written] = 0x2323;
|
||||
|
||||
if (!tests[i].len && tests[i].chars_written)
|
||||
{
|
||||
ok(GetLastError() == ERROR_INSUFFICIENT_BUFFER,
|
||||
"tests[%i]: expected error=0x%x, got error=0x%x\n",
|
||||
i, ERROR_INSUFFICIENT_BUFFER, GetLastError());
|
||||
}
|
||||
ok(len == tests[i].len, "tests[%i]: expected len=%i, got len=%i\n", i, tests[i].len, len);
|
||||
|
||||
if (tests[i].dst)
|
||||
{
|
||||
ok(memcmp(tests[i].dst, tests[i].expected_dst, (tests[i].chars_written + 1) * sizeof(WCHAR)) == 0,
|
||||
"tests[%i]: expected dst=%s, got dst=%s\n",
|
||||
i, wine_dbgstr_wn(tests[i].expected_dst, tests[i].chars_written + 1),
|
||||
wine_dbgstr_wn(tests[i].dst, tests[i].chars_written + 1));
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
static void test_undefined_byte_char(void)
|
||||
|
|
Loading…
Reference in New Issue